每日 arXiv

音频 · 语音 · 音乐 · 声学

每日新爬取的 arXiv 论文，按评分排序。中文摘要帮你快速判断是否值得深读。

2026-08-22
日期2026-08-22
已评分—
均分—
最高—

Daily Papers — 2026-08-22

5 papers on audio, speech, music, and acoustics.

1. Beyond Fresh Starts: Stateful Inference for Streaming ASR in Conversational Voice Agents

Authors: Sameep Chattopadhyay, Alexander Erdmann, Mari Ostendorf

Categories: cs.LG Score: 7.20/10 (Obj:8 Id:7 Ind:8 Comp:8 Eff:7 Nov:6 Rep:8)

  • Strength: 提出免训练的推理时状态管理策略（State Carry-over / Rollback），在 fastconformer-114m 上实现轮首 WER 15–21% 相对降幅，总体 WER 在全部 4 个模型-数据集组合上统计显著优于每轮重置基线，且内存开销不足模型 0.6%。
  • Weakness: 总体 WER 相对降幅在 nemotron-0.6b 上仅约 2–2.5%，未见与”FS+文本历史提示”等轻量基线的对照，RB 阈值依赖验证集调参，且未直接测量对下游语音代理的效用。
  • Full review: Claude Code 全文七公理审稿

Modern voice-agent systems rely on streaming speech recognition models that operate under stringent latency constraints. This study shows that, due to the limited memory constraints of real-time processing, these systems are adversely impacted by conversational phenomena such as long silences and backchannels. While many agentic pipelines mitigate this by resetting state at each turn, this approach discards vital context and impairs performance at turn onsets. We propose two state-management strategies that preserve cross-utterance context to reduce onset errors. In experiments with two state-of-the-art streaming models on two spoken dialogue benchmarks, our best method yields an average of 15-21% relative WER reduction at utterance onsets.

2. Vibrato Matching for Modulation Control and Blending in Sound Mixtures

Authors: Jeremy Hyrkas

Categories: cs.SD, eess.SP | 6 pages, 8 figures, International Computer Music Conference (ICMC) 2026 Score: 6.47/10 (Obj:8 Id:7 Ind:8 Comp:5 Eff:5 Nov:6 Rep:8)

  • Strength: 完整开源的 vibrato matching 算法（源码 + 16 个真实录音示例），把逐谐波与残余谱包络 AM 传递叠加到 vibrato suppression 上，能让 CFT 分离算法在 unison 混合上完全失效，为调制控制与声源融合提供可用的工程工具。
  • Weakness: 全文零量化指标，分离降级仅靠谱图目测，且作者自认无听音实验，核心”听者感知降级”主张缺乏直接证据，也缺少对替代方案与计算成本的对比。
  • Full review: Claude Code 全文七公理审稿

In sound mixtures of more than one musical source, different vibrato patterns act as a cue that multiple sources are present for both human listeners and source separation algorithms. Matching the vibrato patterns of the signals in the mixture reduces the perception of multiple sources, particularly when the sources play in unison. This work introduces the vibrato matching algorithm, which first suppresses vibrato in a target signal and then transfers vibrato from a source signal to the target. An existing vibrato suppression algorithm is combined with a new algorithm for vibrato transfer, which imparts frequency modulation and amplitude modulation to the harmonics of the target signal, and amplitude modulation onto the spectral envelope of the non-harmonic residual component. Examples demonstrate the algorithm’s utility as a vibrato control mechanism and as a tool for blending sound sources. Matching vibrato degrades the performance of source separation algorithms, suggesting a similar degradation in listeners ability to detect the presence of multiple sources.

3. Bulbul: A Dataset for Dialectal Arabic Speech Recognition

Authors: Ahmed Ashraf, Aisha Alansari, Fadel Al Abbas, Nada Almarwani, Samah Aloufi et al.

Categories: cs.CL, cs.AI Score: 6.20/10 (Obj:9 Id:6 Ind:6 Comp:7 Eff:8 Nov:5 Rep:5)

  • Strength: 跨 11 国 10 方言的统一受控数据集（1160 句×10 方言 + 460 句 MSA），含元数据 + 完整录制/验证流程，3 个 ASR 基线（Whisper-large-v3/mHuBERT/Conformer-CTC）的 WER/CER 全报告。
  • Weakness: 数据集/代码/超参数均无可访问链接（GitHub/arxiv 检索 0 命中），基线仅 3 模型无近期 SOTA，且每方言仅 1-9 名说话人未明确 speaker-disjoint 切分，微调实验不可独立复现。
  • Full review: Claude Code 全文七公理审稿

Arabic automatic speech recognition (ASR) faces unique challenges due to diglossia, extensive regional dialect variation, and limited speech resources. Existing speech datasets often focus on single dialects or large-scale broadcast/web data, leading to trade-offs between linguistic diversity and annotation quality. We present BULBUL, a multi-dialect Arabic ASR dataset collected from 275 speakers in 11 Arab countries. BULBUL includes structured dialect and sub-dialect coverage, as well as recordings of classical Arabic and modern standard Arabic spoken by participants in their native dialectal accents to support accent-aware modeling. The quality of the recordings was ensured through a two-level human verification process. We further benchmark a range of recent ASR systems, establishing strong baselines for modern dialectal and accented Arabic ASR.

4. FlowSep 2: Self-Supervised Flow Matching for Language-Queried Audio Source Separation

Authors: Yi Yuan, Xubo Liu, Haohe Liu, Xiyuan Kang, Mark D. Plumbley et al.

Categories: cs.SD | Submission to IEEE/ACM Transactions on Audio, Speech, and Language Processing Score: 5.84/10 (Obj:8 Id:7 Ind:6 Comp:5 Eff:7 Nov:5 Rep:2)

  • Strength: 生成式 LASS 系统性推进，5,650 小时数据 + DiT 缩放 + Self-Flow 语义对齐，AudioCaps FAD 由 2.86 降至 0.88、CLAP A 达 80.5，六基准全面 SOTA。
  • Weakness: 全篇无 SDR/SI-SDR 失真对比，仅 demo 无代码无权重，推理采样成本未报告，核心指标 CLAP 与训练目标存在循环评测风险。
  • Full review: Claude Code 全文七公理审稿

Language-queried audio source separation (LASS) aims to extract target sources from audio mixtures according to natural language descriptions, offering a flexible and scalable interface for audio source separation. However, most existing LASS methods rely on discriminative, mask-based models, which estimate masks from the input mixture. These methods often over-suppress target sounds or fail to fully separate them, especially when multiple sound events strongly overlap in complex acoustic scenes. In this work, we propose FlowSep2, a text-conditioned flow-matching generative model for LASS. Instead of directly predicting a separation mask, FlowSep2 learns to generate the target source representation from Gaussian noise in a latent space, conditioned on both the mixture representation and the text query. Specifically, we employ rectified flow matching with a Diffusion Transformer backbone. We further incorporate Self-Flow, a self-supervised flow-matching paradigm, into our LASS framework. By encouraging semantically structured latent representations under the generative objective, Self-Flow improves the model’s ability to separate target sources according to text queries. Experiments on multiple LASS benchmarks show that FlowSep2 achieves state-of-the-art performance and demonstrates enhanced sound separation results in challenging scenarios with overlapping sound events.

5. SPAR-Hate: Auditor-Guided Multi-Perspective Role Reasoning for Bilingual Hate Speech Parsing

Authors: Yifan Lyu, Dianqing Lin, Xinran Li, Jiaqi Qiao, Xiujuan Xu

Categories: cs.AI | 16 pages, 2 figures. Submitted to EMNLP 2026 Score: 5.58/10 (Obj:8 Id:6 Ind:6 Comp:4 Eff:6 Nov:4 Rep:6)

  • Strength: 结构化双语解析统一框架，同一 14B 骨干下 ZH-main 平均 29.83→32.82、EN-main 29.59→37.51，增益集中于严格联合指标；4B 蒸馏学生在中文两轨反超 14B 教师（ZH-main 33.67 vs 32.82）。
  • Weakness: 核心联合指标绝对水平过低（TAH-H 8.71、EN TA 18.68）且 EN 对比基线均在个位数，无代码/权重发布、TBO 自切 test 划分，13 个超参的四阶段编排多数单元触发仲裁、部分消融贡献为负。
  • Full review: Claude Code 全文七公理审稿

Hate speech research has moved from coarse-grained classification towards structured parsing, where systems jointly identify targets, supporting arguments, and target-level labels. Documents with multiple targets, conflicting local readings, or culturally coded language make these bindings difficult to recover. SPAR-Hate is an auditor-guided multi-perspective role-reasoning framework for bilingual hate speech parsing. It decomposes each document into local focus units, elicits evidence-grounded candidates from Victim, Moderator, and Cultural Bystander perspectives, resolves candidate conflicts under grounding and schema constraints, and reassembles sample-level predictions. Experiments on STATE-ToxiCN and a controlled TBO split show gains across local and API backbones, concentrated on strict joint target-argument-label metrics. Full-test integrated-prompt controls, component ablations, and bounded-arbitration diagnostics identify the contribution of separated perspective generation and arbitration. Structured teacher traces also support training a smaller student model.